Falsifiable Commitment Planning for Self-Correcting Web Agents

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Guangyi Liu, Huan Zhao, Quanming Yao
• Venue: arXiv: cs.AI
• Published: 2026-07-27
• Collected: 2026-07-28
• URL: http://arxiv.org/abs/2607.24167v1
• PDF: https://arxiv.org/pdf/2607.24167v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers propose a new framework called FCPAgent for web agents to ensure they stay on track during long-term planning. The framework represents each plan step as a Falsifiable Commitment Unit (FCU) that includes confirming and falsifying evidence, and a confidence score. Execution is organized in a loop of plan-test-repair. When evidence contradicts a commitment, the system localizes the issue and makes adjustments at the execution, skill, or planning level.

What the Document Says

• The proposed framework is named FCPAgent.
• FCPAgent organizes long-term web agent execution in a plan-test-repair loop.
• Each plan step in FCPAgent is represented as a Falsifiable Commitment Unit (FCU).
• An FCU includes subgoal, confirming evidence, falsifying evidence, and a confidence score.

Key Concepts Mentioned

• FCPAgent
• Falsifiable Commitment Unit (FCU)
• long-horizon web agents
• plan-test-repair loop
• confirming evidence
• falsifying evidence
• confidence score

Claims Made by the Document

• FCPAgent achieves a 13.8% relative improvement in average success over the strongest baseline.

Evidence Used

• On WebArena, FCPAgent achieves a 13.8% relative improvement in average success over the strongest baseline.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Long-horizon web agents often go off track before final failure: a trajectory can remain locally plausible even after the current state, reused skill, or plan assumption no longer supports the user instruction. Existing agents can plan, reflect, or reuse experience, but their plans rarely specify the evidence under which an active step should still be trusted. We propose FCPAgent, a falsifiable commitment planning framework for robust long-horizon web agents. FCPAgent represents each plan step as a Falsifiable Commitment Unit (FCU): a subgoal grounded in a reusable skill, together with confirming evidence, falsifying evidence, and a confidence score. Execution is organized as a plan-test-repair loop. The hybrid commitment testing module checks candidate actions before they modify the browser and checks observations after execution; for efficiency, it combines lightweight evidence matching with LLM-based diagnostic verification. When evidence falsifies a commitment, scope-aware repair localizes the contradiction to the execution, skill, or planning level and revises the smallest adequate part. On WebArena, FCPAgent achieves a 13.8% relative improvement in average success over the strongest baseline, with especially large gains on long-horizon tasks.